Abstract
Background Financial wellbeing is a determinant of mental health but the COVID-19 pandemic and the subsequent cost of living crisis generated by inflation have negatively impacted both aspects.
Methods Data come from Understanding Society. We address long-term (1991-2022) relationships between poor financial wellbeing (PFW), poor financial prospect (PFP) and psychological distress (GHQ-36 >= 9) using a conditional logit model as well as recent trends (2019-2022) using a Latent Growth Modelling (LGM) based on the diagonally weighted least squares (WLSMV) accounting for socio-demographic covariates and measures of deprivation. Multiple imputations were used for non-response.
Results Recent years have seen a surge in GHQ cases and PFP and consistent association between GHQ and PFW (OR=1.31; 95%CI=1.29;1.33) and PFP (OR= 1.13; 95%CI=1.12;1.15) are observed. The association between PFW and GHQ-36 slightly weakened during the pandemic [WLSMV: 0.30 (95%CI= 0.28; 0.33) in 2019; 0.27 (95%CI= 0.25; 0.30) in 2020] but strengthened for PFP [0.22 (95%CI= 0.19; 0.25) in 2019, 0.26 (95%CI= 0.23; 0.29) in 2020]. Those renting and those in the most deprived areas were more likely to report PFW whilst those in the least deprived areas and having a mortgage experiencing greater PFP, both contributing to explain psychological distress.
Discussion Policies implemented during the pandemic might have contributed to partially reduce the association between financial wellbeing and mental health but post-pandemic associations in the context of an early cost of living crisis show an increased risk with those living in the most deprived areas or not owing an accommodation more at risk.
Background
Financial wellbeing is a crucial determinant of mental health. Recent years, punctuated by a two-year-long pandemic and a subsequent and ongoing cost of living crisis, have significantly challenged both aspects. In 2024, the UK is facing both a cost of living crisis and a mental health crisis.
Research on financial wellbeing is sparse even though it has become more and more policy relevant (Brüggen et al., 2017) and most studies usually look at financial wellbeing as an outcome and not as an explanatory variable (Kaur et al., 2021). Among those looking at financial wellbeing as an exposure, the association between poor financial wellbeing and poor mental health is consistent and is irrespective of whether financial wellbeing is measured objectively (e.g., income levels or material deprivation) or subjectively (e.g., self-reported financial wellbeing). A systematic review of 24 cross-sectional studies has found a moderate, yet positive association between financial satisfaction and subjective wellbeing (Ngamaba et al., 2020) but longitudinal studies show stronger associations. For instance, comparing three Australian Cohorts, Butterworth et al. (2009) have demonstrated that financial hardship is much strongly associated with mental health outcomes than other socio-economic characteristics such as employment or education and that this association varies over time as past financial hardship only partially explain later mental health status. Financial hardship correlates with poor mental health and longitudinal measures of such a relationship show that those who have reported financial hardship in the past are more likely to report current mental health problems but mental health problems are greater when financial hardship is reported (Kiely et al., 2015).
Financial wellbeing is not only an individual problem – it can also become a social problem as, when a large group of people faces financial difficulties, this leads to reduced consumption rates and increased reliance on social benefits, potentially affecting the welfare of the society (Brüggen et al., 2017). Beyond an individual perspective focusing on the associations between financial wellbeing and mental health, structural contexts also affect the nature of this relationship. Economic crisis have a true cost in terms of psychological wellbeing (van Hal, 2015) but the role of contextual factors such as inflation or unemployment are not well known (Kaur et al., 2021). Three recent events are of particular interest: the 2008 Great Recession, the 2020-2021 COVID-19 pandemic and the subsequent cost of living crisis fuelled by a massive inflation. The impact of the 2008 Great Recession on (mental) health has been well documented showing that men were more at risk of poor mental health during the crisis and that strong social security systems, particularly in Europe, may have mitigated the effect of Recession (Margerison-Zilko et al., 2016). Other studies have shown that life satisfaction appears to be uncorrelated with GDP growth and the effect of the crisis on overall life satisfaction is small (Mertens & Beblo, 2016). However, specific sub-populations where more at risk to be affected by the crisis with potential long-lasting effects on their mental health in post-recession times (Forbes & Krueger, 2019). By contrast, exposure to economic shock and economic vulnerability – which has somehow increased during the COVID-19 pandemic (Botha et al., 2021) – was associated with higher risks of stress, anxiety and depression (Codagnone et al., 2020). However, even though it was documented that the pandemic was associated with a sharp increase in mental health problems (Patel et al., 2022) and financial inequalities (Blundell et al., 2020; Marmot & Allen, 2020), the associations between both dimensions has not been clearly identified. More recently, the cost of living crisis that was fuelled by inflation due to post-COVID-19 global consumer demand, supply chain disruption and soaring energy priced due to the Russia invasion of Ukraine may has have had an effect on financial wellbeing as inflation leads to higher poverty levels, greater income inequalities, debt problems and other financial difficulties, all of which are associated with worse physical and mental health (Marmot et al., 2013).
Examining the relationship between financial wellbeing and mental health raises two methodological questions. On the one hand, one must account for reverse causation as those with poor mental health are more likely to face financial difficulties because of health care costs (which vary by country), low sick leave benefits or labour market consequences (Swensen & Urban, 2023). In a study using UK data from Understanding Society (Downward et al., 2020) it was shown that (mental or self-reported) health impacts subjective financial situation independently of actual household incomes, demonstrating that improving mental health may improve financial wellbeing. However, the paper suffers from many pitfalls including a non-adjusted household income variable (i.e., not using an equivalence scale to adjust for household composition) and the lack of control for non-income financial variables such as household tenure or savings. Other studies, particularly during the Great Recession, have also indicated that whilst incomes partially explain health, accounting for wealth largely improves the model (López-Casasnovas & Saez, 2020). On the other hand, the correspondence and mismatch between objective and subjective indicators of financial wellbeing has been discussed (Brüggen et al., 2017). On such aspect, it was observed that a decrease in financial resources was associated with increased financial strain during the Great Recession. However, unlike objective measurements, financial strain was a robust predictor of worsening mental health (Wilkinson, 2016). Studies indicates that all measures of financial wellbeing including financial capability, financial distress and financial security lead to the same effect in terms of physical and mental health (Bialowolski et al., 2021) but with varying degrees of association.
Using panel data form Understanding, this article focuses on the UK case. The interventionism of the UK Government during the Covid-19 pandemic has been praised by opposition with its lack of intervention during the following cost of living crisis (Iacobucci, 2022). For instance, the COVID-19 job retention scheme (furlough) has contributed to maintain a large part of the population in employment and to minimize unemployment during the pandemic. This resulted in better mental health levels among the furloughed population (though not as high as for those who kept working), higher compared with those who moved or remained in unemployment during the period (Wels et al., 2022). Similarly, the Credit Holiday scheme was implemented for borrowers to be able to request a delay in repaying their financial or mortgage debts, resulting in mental health benefits among those who used the scheme (Sparkes et al., 2023). Yet, whilst these interventions have mitigated the economic costs of the COVID-19 crisis and financial insecurity for (most) people, population mental health was drastically affected. Social isolation (Mansfield et al., 2023) and home working (Wels et al., 2023) have, for instance, contributed to explain such a trend. The cost of living crisis is a different story. The UK has experienced the highest surge in prices over the past 30 years, exceeding the increase seen after the 2008 financial crisis and surpassing other comparable countries. Mechanisms that explain poor mental health outcomes are mainly related to insecurity to meet basic needs. This comes after a decade of austerity implemented since the Cameron Government in 2010 that led to social security cuts with, for instance, caps on social benefits (Andersen & Reeves, 2022) or a reform of the universal tax credit (Craig et al., 2022). In contrast to the pandemic period, the policy response to the cost of living crisis has been seen as minimal with tax rises and cuts to public spending amidst great uncertainty (Broadbent et al., 2023), leading to a potential public health crisis (The Lancet Regional Health – Europe, 2023).
The study has four main objectives. Objective 1: To examine the association between poor financial wellbeing, financial prospects, and mental health while controlling for socio-demographic, socio-economic, and health-related confounders. Objective 2: To investigate whether the periods of the Covid-19 pandemic and the subsequent cost of living crisis have contributed to an increased likelihood of experiencing poor financial wellbeing and financial pessimism, particularly among the most socioeconomically deprived segments of the population. Objective 3: To assess whether the identified associations between poor financial wellbeing, financial pessimism, and mental health could explain structural shifts observed in mental health trends since 2019. Objective 4: To address whether poor financial wellbeing and financial wellbeing are associated with psychological distress and whether these relationships are distinct.
Data and methods
Understanding Society
We use a full and restricted sample from Understanding Society (USoc). In the full sample, we combine two datasets: the British Household Panel Survey (BHPS) that was a cohort-based study containing 18 yearly waves collated from 1991 (BHPS1) to 2008 (BHPS18) and the UK Household Longitudinal Study (UKHLS) that is still running and contains 13 waves collected in 2009-10 (wave 1), 2010-11 (wave 2), 2011-12 (wave 3), 2012-13 (wave 4), 2013-14 (wave 5), 2014-15 (wave 6), 2015-16 (wave 7), 2016-17 (wave 8), 2017-18 (wave 9), 2018-19 (wave 10), 2019-20 (wave 11), 2020-21 (wave 12) and 2021-22 (wave 13). Although the two datasets can be combined within one single dataset called ‘Understanding Society’ (USoc), sample design is quite different (Fumagalli et al., 2017). BHPS is a cohort-based longitudinal survey following household and individuals living in the UK. The first sample was introduced in 1991 with 5,000 household selected at random within Great Britain (i.e., not Northern Ireland). An additional 1,500 households sample in Scotland and 1,500 in Wales was added in 1999 and an additional 2,000 households sample in Northern Ireland was added in 2000. Specific longitudinal and cross-sectional weights are provided by Understanding Society to correct the sample based on the 1999 and 2000 sample refreshments. By contrast, UKHLS has a clustered and stratified, probability sample of 24,000 households living in Great Britain in wave 1 and a random sample of 2,000 household in Northern Ireland. The sample partially includes members of the BHPS. The sample is not restricted and includes all respondents aged 15 to 100 years of age. The restricted sample only focuses on the recent period taking into consideration waves 10 to wave 13, in other words years 2018-19 (pre-pandemic), 2019-21 (pandemic) and 2021-22 (cost of living crisis). The complete sample is organized in a long format, incorporating non-response and missing data, while the restricted sample is structured in a wide format, beginning at wave 10 (designated as the baseline, i.e., 2019) and tracking respondents across subsequent waves. The complete sample includes 569.653 respondents including 219.968 events from 1991 to 2022. The restricted sample includes 27,707 respondents at baseline (2019).
Data access and ethical statement
The University of Essex Ethics Committee has approved all data collection on Understanding Society main study and innovation panel waves. Participant consent was given during data collection. The overall mechanism for gaining consent for participation in Understanding Society is oral. Participants are sent details about the study in advance letters, information leaflets and are given information by interviewers if taking part in a face-to-face or telephone interview. USoc data can be accessed through the UK data archive portal: https://www.data-archive.ac.uk.
Outcome
We use a binary version of the General Health Questionnaire composed of 36 items (GHQ-36) (Goldberg, 1978) that converts valid answers to 12 questions of the General Health Questionnaire (GHQ) to a single scale and then summing, giving a scale running from 0 (the least distressed) to 36 (the most distressed). These items include concentration, loss of sleep, playing a useful role, capable of making decisions, constantly under strain, problem overcoming difficulties, enjoy day-to-day activities, ability to face problems, unhappy or depressed, losing confidence, believe worthless and general happiness. The binary version (GHQ caseness) is calculated with a cut point at 9 (Holi et al., 2003).
Exposures
We focus on two exposure variables.
The first variable of interest is poor financial wellbeing (PFW) (i.e., how well at managing financially these days?). It is coded on five modalities from 1 (living comfortably) to 5 (finding it very difficult) and is transformed into a binary variable where ‘0’ is attributed to those reporting living comfortably, doing alright or just about getting by and ‘1’ to those reporting finding it quite difficult or very difficult.
The second variable looks at respondent’s poor financial prospect (PFP) for the year ahead (i.e., financial pessimism) and originally contains three categories (better, worse, same). The variable is recoded as binary, distinguishing those reporting that their situation will be worse (coded ‘1’) to those reporting that their situation will be better similar or similar (coded ‘0’).
Adjustment levels and control variables
The models controls for the highest level of education (higher education degree versus no-degree (reference)), the country of residence (distinguishing Scotland, Wales, Northern Ireland, North of England (ref.) and South of England including Greater London) and household composition (couple without child(ren) (ref.), couple with child(ren), single with child(ren) and single), main status of activity (employed (ref.), self-employed, student, retired, on sick leave or on maternity/paternity leave) and the presence of one or more chronic health condition (yes, no (ref.)). Additional variables on financial wellbeing are included in the restricted sample with information on housing tenure (renting, owning with mortgage or owning without mortgage (=reference)), ability to save money on a regular basis (yes, no (ref.)) and an Index of Material Deprivation provided by USoc (IMD) that measures area of residence deprivation across seven domains including income, employment, education, health, crime, access to services and housing environment with a final deprivation score that is the weighted sum of these domains and converted into quintiles (reference: third quintile).
Models
Three different techniques are used.
First, we produce descriptive statistics based on macro-data from 1991 to 2022 that are the percentages GHQ-caseness, poor financial wellbeing, financial wellbeing and a macro indicator of inflation that is the January consumer price inflation index provided by the Office for National Statistics (Office for National Statistics, 2024). We provide correlation coefficients for the time period that is considered and stratify descriptive statistics by gender.
Second, we use conditional logit model (i.e., fixed-effects logit for panel data) using respondents’ ID as a strata and controlling for the year as a fixed effect. The model fits maximum likelihood with a dichotomous dependent variable, calculating the relative likelihood to each group. All stable measured and unmeasured characteristics are controlled and the model only examines within-subject variation, ignoring between-subject variations. We run the model on the full sample and then include a multiplicative interaction term between the exposure and year (as factor). The models are cumulatively adjusted for socio-demographic, socio-economic and health characteristics. All variables are time-varying except age and sex. We provide odds ratios derived from log odds including 95 percent confidence intervals (95%CI). We repeat the models with an interaction between financial variables and gender and years of data collection.
Third, we focus on the restricted sample and apply a Latent Growth Modelling (LGM) technique within the Structural Equation Modelling (SEM) framework (Hox & Stoel, 2014). This approach allows us to construct a latent curve representing GHQ-caseness across multiple time points, capturing an intercept (at wave 10) and a slope that captures subsequent changes over time. The primary aim is to investigate the influence of poor financial wellbeing or poor financial prospect on GHQ-caseness across these four time points. The model adjusts for the different layers of adjustment where each control variable is linked to both PFW and PFP across the different time points but also to the intercept and the slope of the GHQ variables to control for both potential causation issues and indirect effect of the covariates that would not be mediated by PFW and PFP. The covariates are predominantly collected at the baseline due to limit variability observed across the four years, except for the employment status, which varies across time points. Including these covariates allows us to comprehensively explain the variability in both poor financial wellbeing or financial pessimism and the trajectory described by the intercept and slope of the latent GHQ-caseness curve over time. Given the binary nature of the outcome variables, the model uses a Weighted Least Squares Mean and Variance adjusted (WLSMV) estimator that performs better than the logit model (Suh, 2015). The model is run separately for poor financial wellbeing and financial pessimism but a fully adjusted model including both variables is also analysed. Further model specifications can be found in supplementary file S.1.
All analyses are made using the packages ‘survival’ and ‘lavaan’ in R (version 4.2.3).
Weights and missing data
USoc-provided weights are used for all analyses. A cross-sectional weight to ensure sample representativeness is used in the descriptive section and the LGM (wave 10) and a longitudinal weight is used in the fixed effects model. Missing data due to attrition and non-response are imputed using a Random Forest estimator through the ‘Mice’ package in R.
Sensitivity analyses and additional analyses
The study includes several sensitivity analyses and additional checks. First, we stratify descriptive results for the full sample by gender and provide correlation matrices for the variables used (supplementary file S.2). Second, we provide estimates for the multiplicative interaction between poor financial wellbeing and financial pessimism and year in the non-imputed and imputed datasets and replicate descriptive statistics (supplementary file S.4). Third, we replicate the conditional logit based on an extended definition of poor financial wellbeing where the variable is code ‘0’ for those reporting living comfortably or doing alright and ‘1’ for those reporting just about getting by, finding it quite difficult or very difficult (supplementary file S.5). Fourth, we conduct additional analyses using wave 13 as an alternative baseline, progressively including respondents backward through the waves, for the restricted sample to mitigate potential selection bias associated with wave 10 as the baseline (Supplementary file S.7). Finally, we run a full LGM model including both PFW and PFP without controls for the intercept and slope (supplementary file S.9).
Results
Long term trends in mental health and financial wellbeing
Figure 1 shows the percentage of respondents reporting financial difficulties, PFP and GHQ-caseness by year for the full population as well as the consumer price inflation in percentage derived from the Office for National Statistics data on the y-axis. Except for the last few years, there is no straightforward evidence of an association between financial wellbeing and PFP and psychological distress. Over the past three years, we observe a slight increase in PFP with more than 20 percent of the sample reporting that their financial situation is likely to be worse in the coming year in 2022 against 14 percent in 2021 and 12 percent in 2019. Over three years, PFP has increased by eight percentage points. This coincides with the sharp increase in consumer price in 2022. The percentage of respondents reporting financial difficulties has slightly dropped during the pandemic and slightly increased in 2022 at the very start of the cost of living crisis, though not fully reflecting the crisis because data were collected in early 2022. Pearson’s coefficients (supplementary file S.2) confirm what can be observed: neither PFW nor PFP positively correlate with GHQ-caseness when looking at descriptive data. Similarly, no association is observed when introducing a 1-year lag for the GHQ variable. Inflation is strongly associated with PFW (Pearson: 0.51) and PFP (Pearson: 0.64). No difference is observed across genders when looking at the financial variables but the percentage of GHQ-caseness is higher among female respondents.
By contrast, results from the conditional logit model show a positive and statistically significant association between measures of financial wellbeing and mental health. Figure 2 shows the main estimates flowing from the conditional logit where GHQ-caseness is the outcome and PFP and PFW are the exposure variables including four levels of adjustment and interactions with gender in the imputed dataset. Full data from the imputed and non-imputed dataset are in supplementary file S.3. Results show that PFW is a stronger cofounder of GHQ-caseness with Odds Ratios (OR) of 1.34 (95%CI=1.32-1.36) in the unadjusted model and 1.31 (95%CI= 1.29-1.33) in the model adjusting for socio-demographic, socio-economic and health condition variables. Socio-economic factors including the main status of activity marginally reduce the coefficients. The intensity of the relationship is strongly higher among male respondents with OR of 1.39 (95%CI= 1.35-1.42) than in the female sample (OR= 1.26 95%CI: 1.23-1.28). PFP is more often found among the population than PFW but less strongly associated with GHQ-caseness (fully adjusted OR= 1.13; 95%CI= 1.12-1.15). No variation is observed when looking at the different adjustment levels and we observe higher – but not statistically significantly – odds among male respondents. Extra-analyses including a multiplicative interaction between year and the financial variables in the fully adjusted model using the imputed dataset (supplementary file S.4) show that coefficients (in OR) were stable over time with a mean of 1.31 for PFP and 1.13 for PFW. However, whilst no coefficient significantly different from the mean is observed for PFP, higher associations were found for PFW in years 2020, 2021 and 2022.
Sensitivity tests were made using another cut-point for PFW (supplementary file S.5.) with no significant differences across trends and estimates.
Recent trends in mental health and financial wellbeing
Selected estimates from the fully adjusted imputed models are shown in figure 3 both for PFW and PFP. Full results from the LGM for the non-imputed and the imputed datasets and for each adjustment level are shown in supplementary file S.6.
When looking at PFW in the imputed datasets, WLSMV estimators for the unadjusted model are 0.36 (95%CI= 0.37; 0.47) in 2019, 0.28 (95%CI= 0.23; 0.34) in 2020, 0.33 (95%CI= 0.28; 0.39) in 2021 and 0.44 (95%CI= 0.38; 0.49) in 2022 against, respectively, 0.30 (95%CI= 0.28; 0.33), 0.27 (95%CI= 0.25; 0.30), 0.29 (95%CI= 0.27; 0.31) and 0.31 (95%CI=0.28; 0.33) in the fully adjusted model. The relationship between PFW and poor mental health was somehow mitigated during the COVID-19 periods but came back to 2019 level in 2022. When looking at the financial variables associated with PFW, we observe that IMD, housing tenure and capacity to save money are strong predictors of PFW at baseline and over the subsequent waves. In 2019, respondents living in deprived neighbourhoods (IMD 1 and 2) had increased WLSMV of PFW of respectively 0.08 (95%CI= 0.02; 0.15) and 0.08 (95%CI= 0.2;0.15). Those in the least deprived areas (IMD 4 and 5) exhibited lower WLSMV of PFW of −0.04 (95%CI= −0.10; 0.2) and −0.13 (95%CI= −0.19; −0.7). Same associations are observed in 2020, 2021 and 2022 but with slightly higher associations for those in IMD 1 in 2022 (0.13, 95%CI=0.7; 0.20). The capacity for saving money on a regular basis also appears to be a determinant in reporting PFW but the protective effect of savings has reduced in 2022 moving from −0.80 (95%CI= −0.82; −0.74) in 2019 to −0.52 (95%CI=-0.56; −.0.47) in 2022. Renting an accommodation or owing it with an ongoing mortgage is associated with PFW with higher burden for those renting (e.g., in 2022, 0.16 (95%CI=0.10; 0.22) for those owning with mortgage and 0.40 (95%CI= 0.35; 0.46) for those renting). The model also checks for the different relationships between the financial variables and the intercept and slope of the GHQ-caseness variable with no strong associations observed indicating that financial wellbeing captures well the IMD, accommodation type or regular saving capacity.
The employment status is also an important cofounder of GHQ-caseness both directly and via the financial wellbeing route. For instance, those on sick leave at baseline were at higher risk of psychological distress both at baseline (intercept) and in subsequent years (slope) respectively by 0.16 (95%CI= 0.02; 0.29) for the intercept and 0.13 (95%CI= 0.07; 0.19) for the slope but independently of financial wellbeing. The model shows that sick leave is associated with PFW at 0.22 (95%CI= 0.5; 0.40), 0.38 (95%CI= 0.19; 0.57), 0.50 (95%CI= 0.31; 0.70) and 0.43 (95%CI= 0.26; 0.60) between 2019 and 2022 indicating that financial wellbeing explains half of the negative associations between sick leave and psychological distress. The association between unemployment and financial wellbeing is also of interest: coefficients were 0.39 (95%CI= 0.27; 0.50) in 2019, 0.66 (95%CI= 0.53; 0.79) in 2020, 0.59 (95%CI= 0.713; 0.594) in 2021 and 0.48 (95%CI= 0.36; 0.60). The intercept and the slope also show positive coefficients of respectively 0.04 (95%CI= −0.06, 0.11) and 0.05 (95%CI= 0.01; 0.09). No statistically significant association is observed between gender and PFW but it is a strong predictor of the GHQ intercept with higher estimates among the female population (0.23 (95%CI= 0.20; 0.27)) but no significant change over time as the value of the slope is near zero.
Results are slightly different when looking at PFP. The first thing to observe is that PFP is associated with poor mental health but to a lesser extent compared with financial wellbeing. In the unadjusted, associations are 0.29 (95%CI= 0.23; 0.34); 0.32 (95%CI= 0.26; 0.37), 0.33 (95%CI= 0.28, 0.39) and 0.24 (95%CI= 0.19, 0.29) for 2019, 2020, 2021 and 2022. They are 0.22 (95%CI= 0.19; 0.25), 0.26 (95%CI= 0.23; 0.29), 0.26 (95%CI= 0.24; 0.29) and 0.21 (95%CI= 0.19; 0.23) in the fully adjusted model. IMD also shows a different pattern as those in the least deprived areas are more likely to report PFP. For instance, in 2019, those in IMD 5 reported a WLSMV of 0.10 (95%CI=0.04; 0.17) and those in IMD 1 a coefficient of 0.04 (95%CI= −0.04; 0.11). The same type of difference is observed in subsequent waves with respectively −0.08 (95%CI= −0.16; −0.01) and 0.04 (95%CI= −0.03; 0.1) in 2020, −-0.045 (95%CI= −0.12; 0.03) and 0.05 (95%CI= −0.02; 0.11) in 2021 and −0.004 (95%CI= −0.07; 0.06) and 0.039 (−0.02; 0.10) in 2021. Similarly, those renting seem to be less pessimistic compared with those who bought a house with a mortgage. In 2019, the association between renting and PFP was −0.14 (95%CI: −0.20; −0.07) and the association with mortgage was −0.06 (95%CI= −0.12; −0.01). In 2022, those renting are still less pessimist about their financial situation with a coefficient of −0.07 (95%CI= −0.13; −0.01) against −0.01 (95%CI= −0.06; 0.05). As for financial wellbeing, saving capacity is associated with lower PFP across all waves. No association is observed between gender and PFP.
Sensitivity checks were made using a backward sample (see supplementary file S.7). Differences in estimates across models are not substantial with an average of 0.001 difference across estimates in the financial wellbeing fully adjusted non-imputed model, 0.002 in the financial wellbeing fully adjusted imputed model, −0.001 in the PFP fully adjusted non-imputed model and 0.000 in the PFP fully adjusted imputed model.
We also ran a full model including both PFW and PFP (only for the fully adjusted model after multiple imputations). Main estimates are shown in figure 4 with full estimates in supplementary file 8. What can be observed is the strong association between PFW and PFP that stands over the years and has even intensified with respectively 0.31 (95%CI= 0.28; 0.34) in 2019, 0.31 (95%CI= 0.28; 0.33) in 2020, 0.33 (95%CI= 0.30; 0.36) in 2021 and 0.37 (95%CI= 0.35; 0.40) in 2022. As in the previous model, the relationship between PFW and psychological distress has weakened during COVID-19 pandemic [respectively, 0.23 (95%CI=0.21; 0.26), 0.20 (95%CI=0.18; 0.22), 0.21 (95%CI=0.18; 0.23), 0.23 (95%CI=0.20; 0.25)] whilst the relationship between PFP and psychological distress was slightly stronger during the pandemic [respectively 0.15 (95%CI= 0.12; 0.17), 0.20 (95%CI= 0.17; 0.22), 0.20 (95%CI= 0.17; 0.22) and 0.13 (95%CI= 0.10;0.15)]. Sensitivity analyses were made for the same model, not controlling for the associations between the covariates and the intercept and slope (supplementary file S.9). Estimates of the relationships between GHQ caseness and PFW are respectively 0.26 (95%CI=0.23; 0.28); 0.21 (95%CI= 0.19, 0.23); 0.22 (95%CI= 0.20; 0.24) and 0.26 (95%CI=0.23; 0.28) whilst there are respectively 0.17 (95%CI=0.14; 0.20); 0.22 (95%CI= 0.19; 0.25); 0.21 (95%CI= 0.18; 0.24) and 0.12 (95%CI= 0.10; 0.15) for PFP showing low variations compared with the original model.
Discussion
The relationship between mental health and financial requires to examining both at the macro structures that affect individuals and their change over time and at the micro interactions that explain how financial wellbeing translates into poor mental health. What this study demonstrates is that macro associations do not stand as higher rates of poor financial wellbeing or poor financial prospects do not correlate with higher rates of psychological distress at macro level. However, micro-associations constantly show an association across these variables, indicating a strong relationship between poor mental health and financial adversity, which is a good example of ecological fallacy (Robinson, 1950). What is striking is that the intensity of such relationships varies over the years indicating that different contexts lead to different associations.
Focusing on the most recent period that is characterised by the COVID-19 pandemic and a subsequent cost of living crisis, our study reveals that the COVID-19 pandemic was associated with an increase in psychological distress among the population but poor financial wellbeing was less strongly associated with poor mental health during the pandemic than before or after it. This can be partially imputed to the many policy interventions that have been implemented to protect jobs and ensure financial stability throughout the pandemic. These interventions aimed not only to protect populations against the virus itself but also to address a potential economic downturn and the effects it could have on people. These policies were not perfect and the pandemic has certainly exhibited and amplified pre-existing inequalities (Marmot & Allen, 2020) but the implementation of the COVID-19 job retention scheme and the credit holiday have somehow mitigated the economic impact of the pandemic. Consequently, the COVID-19 period saw an increase in cases of psychological distress but economic routes that usually explain psychological distress in normal times have been somehow mitigated. The opposite is true during the early stage of the subsequent cost of living crisis: the lack of intervention has led to a sharpened relationship between psychological distress and poor financial wellbeing, suggesting that the increase in mental health problems after the pandemic can be greatly explained by an increase in financial difficulties.
By contrast, the association between financial pessimism and mental health was stronger during the pandemic and has slightly declined in 2022 (even though the percentage of respondent being concerned about their financial prospects has increased during the period). Poor financial wellbeing and poor financial prospects are indeed two distinct aspects that do not affect people equally but translate, independently, into poor mental health outcomes, although with stronger associations for poor financial wellbeing. One of the reason for this is that the profile of people reporting being pessimist about their economic future is different from those reporting facing financial difficulties. For instance, when looking at housing tenure, we observe that renting an accommodation is associated with higher likelihood of psychological distress compared with fully owning a home without a mortgage and that having a mortgage slightly contribute to explain poor financial wellbeing. Conversely, owing a home with or without a mortgage is associated with higher financial pessimism compared to renting. Renters report poor financial wellbeing whilst owners report greater financial anxiety. Both are likely to report poor mental health, but likelihoods are higher among those actually reporting poor financial wellbeing. The same difference is observed when looking at material deprivation: those in the most deprived areas report greater financial difficulties, while those in the least deprived areas report greater financial anxiety.
This study is not without limitations. First, we purposely use no causal language in this article and did not include arrows in our figures to avoid misinterpretation of the finding but causation remains an issue. Neither conditional logit nor latent growth models fully tackle causation issues. The Structural Equation Modelling Framework (SEM) offers the possibility to control for associations at the intercept and look at the change over time (slope) but the sense of the relationships remains challenging. A second issue is about the nature of the coefficients in the conditional logit and the LGM. Odds ratio can be misleading. If the odds ratio are interpreted as a relative risk, it will always overstate any effect size: the odds ratio is smaller than the relative risk for odds ratios of less than one and bigger than the relative risk for OR of greater than one. They do not approximate well the relative risk when the initiate risk (the prevalence of the outcome of interest) is high (Davies et al., 1998). Similarly, WLSMV are hard to interpret but are the only choice when it comes to binary output within the Lavaan framework. A third issue concerns the use of GHQ-36 and the cut point to make it binary. Although cut point decision may be driven by the research object (Kelly et al., 2008), Fourth, as fieldwork for each Understanding wave takes less than two years, we were unable to control for monthly variations. Wave 10 collection was interrupted at the start of the COVID-19 pandemic, which is why we selected it as the baseline. Both waves 11 and 12 were collected at different time points during the pandemic but we were unable in this study to include information on the different lockdowns. Wave 13 was collected after Covid-19 restrictions ceased, starting in January 2021 and ending in December 2022. For wave 13, data collection timing was at the very early stage of the cost of living crisis. Therefore, this study only addresses the start of the cost of living crisis. Finally, the decision was purposely made to not include individual or household incomes in the survey because the relevance of such a variable is very context-dependent with housing tenure, IMD or wealth playing a major role.
Nevertheless, this article offers some new perspectives on the current state of knowledge. A few studies have focused on the associations between objective and subjective measures of financial hardship or financial strains showing various but consistent degrees of association with mental health or wellbeing. Our study shows that financial wellbeing certainly better captures actual financial strains for, at least, three reasons. Firstly, no empirical study can fully capture the different dimensions of objective financial conditions. Individual and household incomes, social benefits, housing tenure, regular saving, material deprivation, adequation between job and education, cost of transports, energy bills and many other dimensions but also financial knowledge and behaviours (Kaur et al., 2021) play a role and it can be expected that they are better reported by individuals themselves through subjective financial wellbeing measurements than by survey items that are limited by nature. Secondly, our study shows that financial wellbeing is clearly distinct from financial prospects demonstrating that respondents can distinguish their own condition from economic uncertainty and, therefore, the adequacy of such a measurement. Interestingly, we found that those in the most deprived areas report poor financial wellbeing whilst the least deprived report financial pessimism. Finally, there is little evidence in the literature that mental health would bias self-report financial wellbeing but the use of longitudinal data and control for baseline values remains necessary.
Funding
JW is the PI of the Horizon Europe ERC Starting Grant UHealth (Trade Unions and Workers’ Health), the PI of the “Negotiating Health: addressing the role of trade unions in explaining workers’ physical and mental health” (NegHealth) Incentive Grant for Scientific Research funded by the Belgian National Scientific fund (FNRS) and the PI of the “Trade unions, job satisfaction and workers’ health: an international comparison of panel data” Joint Usage and Research Center Program at the Institute of Economic Research, Hitotsubashi University, Japan.
Data access and ethical statement
The University of Essex Ethics Committee has approved all data collection on Understanding Society main study and innovation panel waves. Participant consent was given during data collection. The overall mechanism for gaining consent for participation in Understanding Society is oral. Participants are sent details about the study in advance letters, information leaflets and are given information by interviewers if taking part in a face-to-face or telephone interview. USoc data can be accessed through the UK data archive portal: https://www.data-archive.ac.uk.
Conflict of interest
The authors report no conflict of interest. JW is a member of the Belgian Health Data Agency (HDA) user committee.
Data Availability
The study uses Understanding Society data that can be accessed through the UK data archive portal: https://www.data-archive.ac.uk.
Supplementary files
Supplementary file S1. LGM specifications
Supplementary file S2. Stratification by gender and correlation matrices Supplementary file S3. Conditional logit main estimates
Supplementary file S4. Conditional logit with year multiplicative interaction (fully adjusted model only)
Supplementary file S5. Sensitivity analyses of poor financial wellbeing Supplementary file S6. LGM results
Supplementary file S7. Sensitivity analyses for LGM results using backward sampling
Supplementary file S8. LGM results including both poor financial wellbeing and financial pessimism.